Professional Documents
Culture Documents
Healy 1987
Healy 1987
North-Holland
Krishna G. PALEPU
Ilaruard Busmess School. Boston. MA k?163. (I’sA
This paper examines the effect of accounting procedure changes on cash salary and bonus
compensation to CEOs. We estimate whether there is an adjustment to the statistical relation
between compensation and corporate earnings following changes that lower earnings (FIFO to
LIFO inventory valuation) and that raise earnings (accelerated to straight-line depreciation). The
results indicate that (1) subsequent to these changes salary and bonus payments are based on
reported earnings, rather than earnings under the original accounting method, and (2) the
potential compensation effect of the changes is small compared to the elfect of economy- or
industry-wide changes in compensation.
1. Introduction
In many large U.S. corporations, executives’ remuneration is explicitly
linked to reported earnings. For example, in 1980 more than 90 percent of the
1000 largest U.S. manufacturing companies used some form of earnings-based
compensation plan [see Fox (198O)J.l Recent studies in accounting hypothesize
that these earnings-based plans induce managers to select income-increasing
accounting procedures, or impede their selecting income-decreasing proce-
dures, since compensation is imperfectly adjusted for the effect of alternative
*We wish to thank Andrew Christie, Dan Collins (the referee). Linda DeAngelo. Michael
Jensen, Bob Kaplan Rashad Abdel-khahk, Pat O’Brien, Rick Ruback. Ross Watts, Jerry
Zimmerman, and the participants at the Stanford Summer Workshop and the MIT Accounting
Workshop for their helpful comments on earlier drafts of this paper. We are also grateful to Bob
Holthausen and Richard Rikert for letting us use their data bases of changes in accounting
procedures.
‘The most popular schemes included bonus and performance plans. Bonus plans typically
award managers cash payments if certain annual earnings targets are attained: performance plans
award managers the value of performance units or shares in cash or stock if certain long-term
(three- to five-year) earnings targets are achieved. For a more detailed description of bonus plans,
see Healy (1985). For a description of performance plans. see Smith and Watts (1982) and Larcker
(1983).
accounting methods. [See Holthausen and Leftwich (1983) and Watts and
Zimmerman (1986) for a summary of these studies.]
This paper examines the statistical relation between cash salary and bonus
remuneration to chief executive officers (CEOs), and corporate earnings. We
test whether there is an adjustment to this statistical relation subsequent to an
accounting procedure change. Two forms of adjustment are considered: (1)
those that transform reported earnings under the new accounting method to
earnings under the original method, and (2) adjustments to the parameters of
the relation that offset the effect of the accounting change. We also examine
the effect of an accounting change on CEOs’ salary and bonus awards if no
adjustment is made to either reported earnings or parameters of the compensa-
tion-earnings relation.
We investigate two accounting method changes: from the FIFO to LIFO
inventory method, and from accelerated to straight-line depreciation.2 These
are selected for the following reasons. First, they have a large effect on
reported earnings. Second, FIFO to LIFO switches typically decrease reported
earnings, whereas changes from accelerated to straight-line depreciation usu-
ally increase reported earnings.j By investigating both income-increasing and
income-decreasing changes we are able to increase the power of our tests. Our
inventory sample comprises 52 test firms that changed from the FIFO to
LIFO inventory method and 50 control firms with no inventory or deprecia-
tion method changes. The depreciation sample contains 38 test firms that
changed from accelerated to straight-line depreciation and 37 control firms.
We conclude that (1) subsequent to the accounting changes, cash salary and
bonus awards are based on reported earnings rather than ‘as-if’ earnings; (2)
the parameters of the compensation-earnings relation change for both the test
and control firms subsequent to the accounting changes: and (3) the potential
impact of the method changes on salary and bonus payments is small relative
to economy-wide changes in compensation.
A number of earlier studies have examined the relation between inventory
and depreciation accounting method changes and executive compensation.
Their results are mixed. For example, Holthausen (1981) investigates whether
the cross-sectional variation in firms’ abnormal stock price performance at the
date of a change from accelerated to straight-line depreciation is related to the
existence of an earnings-based compensation plan. No relation is found.
‘A number of earlier studies that investigate the stock price reaction to accounting changes also
focus on these two changes, They include Kaplan and Roll (1972) and Holthausen (1981) in the
case of depreciation changes, and Sunder (1973,1975), Biddle and Lindahl (1982) and Ricks
(1982) in the case of inventory changes.
‘Changes from FIFO to LIFO decrease reported earnings if input prices are rising and physical
inventory levels are not depleted. Changes from accelerated to straight-line depreciation increase
reported earnings if nominal investments in new depreciable assets are increasing.
P. Heu(v, S. Kung and K. Palepu, Accounting procedure changes und CEO compemut~on 9
However, as Holthausen and Leftwich (1983) point out, this test lacks power
because the compensation effects of accounting changes are likely to be small
relative to the variability in stock prices and, it is difficult to identify the event
dates and specify investors’ expectations. Hagerman and Zmijewski (1979)
investigate whether companies that have earnings-based compensation plans
are more likely to use the FIFO inventory method and straight-line deprecia-
tion than companies with no earnings-based plans. They find no relation
between the cross-sectional variation in inventory methods and the existence
of a compensation plan. However, there i s 0 !veak positive relation between
the probability that a company uses straight-line depreciation and the prob-
ability that it has an earnings-based compensation plan.
Abdel-khalik (1985) estimates a cross-sectional regression of salary and
bonus compensation on earnings for a treatment sample of 88 companies that
changed to LIFO in 1974, and a control sample of 88 companies that
remained on FIFO in that year. He estimates differences in the fixed compo-
nent of compensation and the elasticity of compensation to reported earnings
between these two samples in the two years prior to, the year of, and the year
following the accounting change. In the two years prior to the inventory
change the compensation parameters do not differ between the treatment and
control groups. In the year of the change, the fixed component of compensa-
tion is higher for the control group and the elasticity of compensation to
reported earnings is higher for the treatment group. These differences persist
when reported earnings are replaced by as-if FIFO earnings for the treatment
group. Abdel-khalik’s findings are therefore mixed.
Our test design differs from that of Abdel-khalik in three ways. First, we
examine two accounting changes, one that typically increases earnings (a
change to straight-line depreciation) and another that decreases earnings (a
change to LIFO). Second, our tests examine earnings and compensation data
for as many as ten years following an accounting change. Third, we use a
time-series rather than a cross-sectional approach to estimate the relation
between compensation and earnings. Murphy (1985) points out that studies
that regress compensation on some index of performance across executives at
a particular point in time are likely to be misspecified. For these regressions,
‘the exclusion of individual-specific factors, such as education and training,
perceived ability, performance in previous jobs, firm size, etc., will lead to an
omitted variables problem, reflecting factors that are fixed for an executive
over time but vary across executives at a point in time’ (p. 22). Correlations
between the independent variables in a cross-sectional regression and these
omitted variables will bias the estimated coefficients. There is some evidence
that Abdel-khalik’s findings suffer from this form of misspecification because
when firm size is included in his cross-sectional regressions, the fixed compo-
nent of bonus and salary awards and the elasticity of compensation to
earnings no longer differ between the treatment and control groups.
10 P. Heu(r’. S. Kong and K. Pdepu, Accounting procrdurechanges and CEO compensu~mn
Table 1
Number of test firms changing to LIFO or straight-line depreciation by year in the period
1967-1976.
Number of firms
Year of Number of firms changing to straight-
accounting change changing to LIFO line depreciation
1967
1968
1969 _
1970 1
1971 1
1972 0
1973 0
1974 36
1975 10
1976 4
Total 52
Table 2
Number of test firms changing to LIFO or straight-line depreciation in the period 196771976 hy
2-digit industry
Number of firms
2-digit changing to
SIC Number of firms straight-line
code Industry changing to LIFO depreciation
20 Food products
21 Tobacco manufacturers
22 Textile mill products
23 Apparels and fabric products
26 Paper and paper products
28 Chemicals
29 Petroleum refining
30 Rubber and misc. plastic products
31 Leather and leather products
32 Stone, clay, glass and concrete products 2
33 Primary metal industries 9
34 Fabricated metal products, except
machinery and transportation equipment 1
35 Machinery, except electrical 4 9
36 Electrical and electronic machinery.
equipment and supplies 4
37 Transportation equipment 5
38 Measuring, analyzing. and controlling
instruments. photographic. medical and
optical goods, watches and clocks 1
39 Miscellaneous manufacturing industries 1
51 Wholesale trade-nondurable goods 1
53 General merchandise stores 3 1
54 Food stores 2 1
59 Miscellaneous retail stores 1
Total 52 38
12 P. Heui,~. S. Kmg md K. Palepu, Accoun~rng procedure changes md CEO con~permtron
Table 3
Summary statistics for average annual executive cash salary and bonus, and corporate earnings for
sample firms.”
Mean
standard First Third
Mean deviation quartile Median quartile
Inventory:
Test sample 116,216 47,786 26.312 52.233 89,694
Control sample 73.078 30.495 5.747 36.067 76,179
Depreciation:
Test sample 52,308 21.293 15,032 3 1,247 67.877
Control sample 57,736 29,881 14.007 34,395 49.920
“These results are for the cross-sectional distribution of time-series averages of annual executive
cash salary and bonus, and corporate earnings for each firm in the inventory and depreciation
samples, Each time-series contains between 14 and 21 observations.
‘Both executive salary and bonus, and corporate earnings are in thousands of CPI-deflated
(1967 = 100) constant dollars.
‘The test sample comprises 52 firms that change to LIFO in the period 1970-1976 and 38 firms
that change to straight-line depreciation in the period 1967-1974. Each test Arm has a matched
control firm with the same 2-digit SIC code and with no inventory or depreciation method
changes ten years before and after the year of the test firm’s method change. Control tirms are
found for 50 of 52 inventory test firms and 37 of 38 depreciation test firms.
available. Compensation and corporate earnings data are deflated by the CPI
to 1967 dollars and time-series averages of these variables estimated for each
company. A summary of the cross-sectional distribution of these time-series
averages is presented in table 3. The median average salary and bonus in 1967
constant dollars is $194,000 for the inventory test sample and $168,000 for the
inventory control sample. The median values for the depreciation test and
control samples are $178,000 and $165,000, respectively. The median average
corporate earnings in 1967 dollars is $52,233,000 for the inventory test sample
and $36,067,000 for the inventory control companies. The median values for
the depreciation test and control firms are $31,247,000 and $34,395,000.
The earnings effects of the inventory and depreciation changes are collected
for the test firms from the financial statement footnotes for years following the
change. Companies that use the LIFO inventory method report the current
replacement value of inventory. This value approximates the FIFO inventory
value and is used to calculate the difference between reported LIFO income
and income that would have been reported had the company continued to use
FIFO.
P. ffea!v. S. Kung and K. Palepu, Accounting procedure changes und CEO c‘ompetwutroll 13
Table 4
Summary statistics for earnings effects of inventory and depreciation method changes as a
percentage of ‘as-if’ earnings for the year of the accounting method changes and the subsequent
ten years.”
“Our depreciation sample includes six companies that changed to straight-line depreciation
after 1970. For these companies the deferred tax items after 1981 refect Accelerated Cost
Recovery System (ACRS) rates. ACRS decreased the depreciable lives of fixed assets relative to
previous accelerated methods, implying that subsequent to 1981 our adjustments for these
companies are not strictly comparable to accelerated depreciation used for reporting purposes.
However, this change affects at most three years of data. and is unlikely to alter our conclusions.
‘None of the companies in our sample reports the deferred tax effect for depreciation prior to
1971. Our as-if earnings series is therefore incomplete for companies that change to the straight-line
method prior to this date. These years are treated as missing observations in the empirical tests.
Table 4 reports statistics on the earnings effect of the change to LIFO as a
percentage of earnings under FIFO. During the year of change, the median
earnings reduction from the inventory policy change is 19.3 percent. Time-series
averages of the earnings effect of the inventory change are estimated for each
firm for the first five years (years 1 to 5) and the second five years (years 6 to
10) following the LIFO change. The cross-sectional median of these averages
is - 8.3 percent for the first five years and - 1.1 percent for the second five
years. The percentage effect of the depreciation change on earnings calculated
using accelerated depreciation is also reported in table 4. The median increase
in earnings from the depreciation change is 9.6 percent in the year of the
change. The cross-sectional median of the average earnings effect for the first
live years subsequent to the depreciation change is 10.9 percent, and for the
second five years is 7.7 percent.
COMP, = salary and cash bonus paid to chief executive offtcer (CEO) during
year t in 1967 constant dollars,
EARN, = accounting earnings before extraordinary items for the firm during
year t in 1967 constant dollars,
D,, = 1 if individual i was CEO of the firm during year t, 0 otherwise,
n zz number of individuals who held the position of CEO of the tirm
during the sample period,
p.a, = firm-specific parameters, to be estimated using time-series data on
compensation and earnings (i = 1,. . , n).
The above compensation model is estimated separately for each firm in the
“This model is hirnilar to a model presented in Murphy (19X5). However. Murphq uses
abnormal stock performance as 3 proxy for management’s performance, and conbtralns the slope
cocfiicicnt (8) to be constantacrossfirms. For our sample, we rqect the hypothesis that the slope
uxHiclent i> equal XKM firms.
P. Heu!v, S. Kung und K. Palepu, Accounting procedure changes und CEO compensatim 15
sample. There are three features of the model that are worth noting. First, the
model is estimated in a logarithmic form since there is some evidence that
power transformations perform better than linear regressions in estimating
relations between compensation and measures of performance [see Boyes and
Schlagenhauf (1979)]. In addition, prior studies have typically used log
transformations [e.g., Murphy (1985) and Abdel-khalik (1985)]. The use of a
logarithmic form, therefore, makes our results comparable with the findings of
these studies.’
A second feature of the compensation model is that the intercept term (or
the fixed component of compensation), (Y,is allowed to vary across executives.
This enables us to account for differences in manager-specific factors such as
age, ability and education. The elasticity of compensation to earnings, /?, is
assumed to be firm-specific.
The third feature of the model is that the compensation variable is rep-
resented by the salary and bonus payments to the chief executive officer. This
variable excludes several components of compensation, such as performance
awards that are contingent on earnings, and stock option compensation,
because disclosures of these awards are frequently incomplete. Their omission
limits the conclusions of the study since compensation committees could
conceivably adjust these awards, rather than short-term earnings-based awards,
to offset the effect of an accounting method change. However, it is worth
noting that bonus and salary comprise a non-trivial proportion of executives’
total remuneration. For example, in Murphy’s study of 461 executives from
1964 to 1981, salary and bonus account for an average 80 percent of total
remuneration.
The compensation model in eq. (1) posits that management compensation is
in part determined by contemporaneous accounting earnings.x A change in the
rules used to compute accounting earnings will therefore affect executives’
earnings-based compensation unless the compensation committee adjusts for
the effect of the accounting change. The committee can insulate salary and
bonus compensation from the effects of an accounting rule change in three
ways. First, it can continue to use earnings computed under the pre-change
accounting rules. In other words, reported earnings under the new accounting
rules are adjusted for the effect of the change and the compensation-earnings
‘If earnings are negative, we assume their log value is zero. Twenty of the 90 companies in our
sample (29 company-years) are adjusted in this way. The effect of this assumption is to limit
managers to receive earnings-dependent compensation only when their company earns profits.
consistent with the option characteristics of most bonus contracts [see Healy (1985)]. We also
estimate mode1 (1) in linear form, allowing earnings to be negative. and constraining negative
earnings to zero. Our conclusions are not sensitive to this adjustment to earnings, or to the
logarithmic transformation.
‘We also test whether lagged earnings are related to salary and bonus awards since salary
adjustments could be based on prior years’ earnings. We find no evidence that lagged earnings are
related to bonus and salary awards.
16 J? Ifrrr~~:S. Kung and K. Pulepu, Accountingprocedure changes und CEO compensation
relation in eq. (1) is applied to the adjusted earnings numbers. Second, the
committee can use reported earnings without any adjustment for the account-
ing change but can modify the parameters (Y and /3 in eq. (1) such that, on
average, the compensation awarded is unaffected by the accounting change.
The tests described below examine whether compensation is insulated from
the effects of accounting changes in either of these two ways. The third way for
the compensation committee to offset the effect of an accounting change on
CEOs’ remuneration is through adjustments to stock-based compensation. As
noted above, we do not collect data on stock-based awards to CEOs for our
sample and therefore do not test this hypothesis.
where
t, = t-statistic for firm j associated with the estimate of the parameter (/3 or
A)*
k, = degrees of freedom in regression for firm j,
N = number of firms in the sample.
The t-statistic for firm j is distributed Student-r with a variance of k,/( k, - 2).
Under the Central Limit Theorem, the sum of the standardized t-statistics is
normally distributed with a variance of N. The Z-statistic for each of the
parameters is therefore a standard normal variate under the null hypothesis
that the parameter (/3 or h) is not significantly different from zero.
A second and equivalent test compares the explanatory power of eq. (2)
with the following restricted form of that equation:
Eq. (3) restricts X in eq. (2) to be equal to zero. An F-statistic is used to test
whether this restriction results in a significant reduction in the explanatory
power of eq. (2):
SSR, - SSR,
F,=
SSRJk, ’
where SSR, and SSR 2 are the sums of squares of residuals of eqs. (3) and (2).
respectively, and k, is the degrees of freedom in regression eq. (2) for firm ,j.
An F-statistic is computed for each firm in the sample. The significance of
the sample distribution of F-statistics is tested by the following statistic:
.Y
x2= C -2lnp,,
/=1
‘For a detailed discussion of both these tests. see Christie (19X6). Roth the tests discussed here
are based on the sample distribution of the parameter estimates. In using these tests. it is assumed
that the parameters are independent across the firms in the sample. Further discussion of this
assumption is deferred to later in the paper.
18 P. H&y, S. Kung and K. Pulepu, Accountrng
procedure changes and CEO cmpmsation
Table 5
Summary of estimated coefficients for regression tests of the hypothesis that reported earnings are
adjusted to FIFO earnings for compensation purposes following a change to the LIFO inventory
method.a
ln(COMP,,)=~a,D,,+Pln(A~ARN,)+Xln + e,.h
,=I
“These results are for the cross-sectional distribution of time-series regressions for 52 firms that
changed to LIFO in the period 1970-1976. Each time-series contains between 14 and 21
observations.
bCOMP, = salary + bonus for CEO in year I: AEARN, = as-if earnings in year t computed
using the FIFO inventory method; REARN, = reported earnings in year t, computed using FIFO
before and LIFO after the accounting change; D,, = 1 if individual 1 is CEO of the firm in year t
and 0 otherwise; and n = number of individuals who held the CEO position in the sample period.
‘Under the null hypothesis each Z-statistic is distributed unit normal.
dSignificant at the one percent level using a two-tailed test.
‘Significant at the one percent level using a one-tailed test.
Table 5 presents regression results for the test sample of 52 firms that
changed from FIFO to LIFO. lo.11 The estimated coefficient p on the adjusted
earnings variable (AEARN) is positive for 92 percent (48 of 52) of the sample
firms. A binomial test indicates that there is a higher proportion of positive
coefficients than expected by chance at the one percent significance level.12
“‘The control sample firms are not analyzed in this section because we do not have information
on adjusted earnings for these firms.
“Twenty-three of the 52 company regression residuals exhibit serial correlation. We use a
Cochran-Orcutt transformation for these companies [see Theil (1971) for a description of this
technique]. The results do not change significantly. The results reported in table 5 arc therefore for
the unadjusted estimates.
‘*See Siegel (1956) for a description of the binomial test. The test assumes that the coefficients
are cross-sectionally independent. As noted above, further discussion of this assumption is
deferred until later in the paper.
The Z-statistic, which tests the joint significance of the AEARN coefficients for
the 52 sample companies, is significant at the one percent level, enabling us to
reject the hypothesis that there is no significant relation between accounting
earnings and top management bonus and salary compensation.
The estimated coefficient h for the variable representing the effect of the
change to LIFO (REARN/AEARN) on executives’ compensation, is positive
for 71 percent (37 of 52 firms) of the sample. A binomial test rejects the
hypothesis that the proportion of positive coefficients is equal to that expected
by chance at the one percent level. The sample Z-statistic, testing the collec-
tive significance of the X coefficients for the 52 firms in the sample. is
significant at the one percent level. i3 This evidence is inconsistent with the
hypothesis that CEOs’ salary and bonus compensation is based on earnings
adjusted for the effect of the inventory method change (FIFO earnings).
Finally, the X’-statistic shown in table 5 aggregates the probabilities associ-
ated with the F-statistics for all firms in the sample. The F-statistic compares
the explanatory power of a regression in which X is restricted to be zero with
one in which it is not. The x’-statistic for the 52 firms in the sample is 184.5
and is significant at the one percent level. This evidence, consistent with that
indicated by the Z-statistic discussed above, suggests that salary and bonus
compensation is related to reported earnings, unadjusted for the effect of the
change to LIFO. There is no evidence that, for compensation purposes,
reported earnings are adjusted to offset the effect of the change in accounting
policy for valuing inventory.
The results for the 38 test firms that changed from accelerated to straight-line
depreciation are similar to the results for the inventory sample. Table 6 reports
a summary of the ordinary least squares estimates of model (2). The estimates
of the earnings coefficient /3 are positive for 84.2 percent (32 of 38) of the
sample firms. A binomial test indicates that this number of positive estimates
is significantly greater than that expected by chance at the one percent level
using a two-tailed test. The Z-statistic, which tests the sample significance of
the estimates of /?, is also significant at the one percent level.
The estimates of X, the coefficient on the earnings variable that captures the
effect of the accounting change, are positive for 60.5 percent (23 of 38) of the
companies. The Z-statistic, which tests the significance of the sample estimates
of h, is significant at the one percent leveli Consistent with this, the
Table 6
Summary of estimated coefficients for regression tests of the hypothesis that reported earnings are
adjusted to accelerated depreciation earnings for compensation purposes following a change to the
straight-line depreciation method.”
In(L.OMP,)=~a,D,,+Pln(AEARh;)~hln + F,.h
,=I
“These results are for the cross-sectional distribution of time-series regressions for 38 firms that
changed to straight-line depreciation in the period 1970-1976. Each time-series contains between
14 and 21 observations.
‘COMP, = salary + bonus for CEO in year t; AEARN, = as-if earnings in year t, computed
using the accelerated depreciation method; REARN, = reported earnings in year 1, computed
using accelerated depreciation before and straight-line depreciation after the accounting change;
D,, = 1 if individual i is CEO of the firm in year t and 0 otherwise; and n = number of individuals
who held the CEO position in the sample period.
‘Under the null hypothesis each Z-statistic is distributed unit normal.
dSignificant at the one percent level using a two-tailed test.
‘Significant at the one percent level using a one-tailed test.
xz-statistic, which tests whether the accounting change effect variable provides
an increase in explanatory power of the sample regressions, is significant at the
one percent level. This evidence is inconsistent with the hypothesis that
compensation committees continue to use earnings based on accelerated
depreciation in determining top management compensation after a change to
straight-line depreciation for reporting purposes.
In summary, our results for both inventory and depreciation samples
indicate that there is a significant relation between top executive salary and
bonus compensation and reported accounting earnings. The results indicate
that when the two accounting policies are changed, compensation committees
do not adjust reported earnings for the effects of the accounting changes.
One potential limitation to our findings is that the statistical tests used to
aggregate regression results across companies assume the sample observations
to be independent. Since the inventory and depreciation changes are clustered
in time and concentrated in several industries, this assumption may be
violated. To investigate cross-sectional dependence, we estimate the cross-sec-
tional correlations of the residuals from model (2) for 42 companies in the
P. Heu!v, S. Kang and K. Palepu, Accounling procedure changes and CEO compensation 21
inventory sample with complete data available from 1966 to 1980, and 24
companies in the depreciation sample with data available from 1962 to 1976.
There are 861 pairwise residual correlation coefficients for the 42 inventory
companies and 276 coefficients for the 24 depreciation companies. The mean
correlation coefficient is only 0.0102 for the inventory sample and 0.0193 for
the depreciation sample, neither significantly different from zero. We thus find
no strong evidence of cross-sectional dependence.
where DUM, is 0 in years prior to the accounting method change and 1 after
the change, & and & are the adjustments to the intercept and slope
parameters of the compensation model following the accounting change, and
D,,, REARN,, (Y, and /3i are as defined in eq. (2).
22 P. Heu!v, S. Kung and K. Puiepu, Accounttng procedure changes und CEO compensation
Under the null hypothesis that the fixed component of compensation and
the elasticity of compensation to earnings are unaltered following an account-
ing change, both & and & are expected to be zero. We also estimate eq. (4)
for control firms using the same dummy variables (DUM,) as were used for
the matched test firms. If there are no economy- or industry-specific changes
in the parameters of the compensation model, & and & for the control firms
are expected to be zero.
We use a chi-square test to examine the hypothesis that & and & are
jointly zero. To compute the &i-square statistic, we estimate the following
restricted form of eq. (4) for each firm in the sample:
Using the residuals of eqs. (4) and (5) the following statistic is computed for
each firm:
F = (SSR,- SSR,)/2
J SSR,/k, ’
where SSR, and SSR, are the sums of squares of residuals for firm j from
eqs. (5) and (4), respectively, and k, is the number of degrees of freedom in
regression (4) for firm j. The sample distribution of the above F-statistics is
used to test their significance by computing the chi-square statistic discussed
in section 3.2.
where 2, and p^, are estimates from eq. (4) for the log of fixed compensation
and the elasticity of compensation to earnings prior to the accounting change,
and COMP,, D,, and AEARN, are defined in eq. (1). Expected compensation
P. Heu!v. S. Kung und K. Palepu, Accounting procedure chunges and CEO compensation 23
is therefore the antilog of the right-hand side of eq. (6). Expected compensa-
tion under eq. (6) is estimated for each test firm for years subsequent to the
accounting change and percentage prediction errors are computed as follows:
PE = COMP, - E( COMP,)
f x 100. (7)
COMP,
where r1 is the first year following the accounting method change and r2 is at
least or + 5. and at most 71 + 10, depending upon data availability.
Under the null hypothesis that the compensation committee adjusts the
compensation-earnings relation to offset on average the salary and bonus
effect of the accounting change, the time-series average percentage prediction
error for each firm is zero. If the compensation committee uses reported
earnings to award earnings-based compensation and does not adjust the model
to offset the effect of the accounting change, the average percentage prediction
error is expected to be negative for inventory test firms and positive for
depreciation test firms. A Student-t test is used to evaluate the significance of
the cross-sectional mean of the average percentage prediction errors.
We also calculate prediction errors for the inventory and depreciation
control firms to examine changes in salary and bonus awards induced by
industry- and economy-wide factors. The control firms do not make changes in
inventory and depreciation methods and therefore do not have adjusted
earnings series. We substitute reported earnings for adjusted earnings in eq. (6)
and estimate predicted compensation and percentage prediction errors for
each control firm during the same calendar years used to predict compensation
for its matched test firm. If there are economy- or industry-specific changes in
compensation that coincide with but are unrelated to the accounting changes,
the time-series average prediction error for the control firms is non-zero.
We compare the average percentage prediction error for each matched pair
of test and control firms and use a Student-t test to evaluate the significance of
differences between the samples. If the compensation committee adjusts the
model parameters to offset the effect of an accounting change on earnings-based
compensation, we expect no difference in average prediction errors between
24 P. Hea(v. S. Kung rend K. Pulepu. Accounting procedure change.7 crnd CEO compensation
EFF = - E(COMP,,)
E(COMP,,)
t COMP,
x 100,
where COMP, is the actual salary and bonus awarded in year t, and E( COMP,,)
and E(COMP,,) are expected compensation in year t based on reported
earnings and adjusted earnings, respectively. E(COMP1,) and E(COMP,,) are
computed as the antilogs of the right-hand side of the following two equa-
tions:
where REARN, is reported earnings, AEARN, is adjusted earnings, &, and p^,
are estimates from eq. (4), and D,, is a dummy defined in eq. (4). The
percentage effects of the inventory and depreciation accounting changes on
salary and bonus awards are estimated for each test firm for years subsequent
to the method changes.
Table 7 presents a summary of the estimates of eq. (4) coefficients for the 50
matched pairs of firms in the inventory test and control samples. Parameters
P. HeaJv, S. Kang und K. Palepu, Accounting procedure changes and CEO compensution 25
Table I
Summary of estimated coefficients for regression tests of the hypothesis that the parameters of the
compensation-earnings relation change following a change to the LIFO inventory method.”
Test sumplec
“These results are for the cross-sectional distribution of time-series regressions for 50 test firms
and 50 control firms. Each time-series contains betwen 14 and 21 observations.
hCOMP, = salary + bonus for CEO in year 1; REARN, = reported earnings in year t; D,, = 1 if
individual i is CEO of the firm in year I and 0 otherwise; n = number of individuals who held the
CEO position in the sample period; and DUM, = 0 prior to the year of the inventory change and
1 thereafter.
‘The test sample comprises 50 firms that change to LIFO in the period 1970-1976. Each test
firm has a matched control firm with the same 2-digit SIC code and with no inventory or
depreciation method changes ten years before and after the year of the test firm’s method change.
Control firms are found for 50 of 52 inventory test firms.
dThese distributional statistics are for the differences in coefficients between matched pairs of
test and control firms. The Z-statistic tests the significance of the sample mean difference for each
coefficient.
‘Significant at the one percent level using a two-tailed test.
‘Significant at the one percent level using a one-tailed test.
26 P. Heu!v, S. Kung und K. Pulepy Accountrng procedure changes und CEO compensation
& and & in the regression are used to examine whether the fixed component
of compensation and the elasticity of compensation to earnings change subse-
quent to a change from FIFO to LIFO. The estimated values of & are positive
for 58 percent of the test firms and the mean value is 1.2497. The percent of
positive coefficients for the control firms is 50 percent and the mean of the
coefficients is -0.2296. Forty-six percent of the estimated values of & are
positive for the test firms and the mean estimate is - 0.0944. The correspond-
ing values for the control sample are 56 percent and 0.0317. The x*-statistics,
which test the hypothesis that & and & are jointly zero, are significant at the
one percent level for both the test and control firms. Thus, there is some
evidence of a structural change in the compensation-earnings relation for the
inventory sample.
An examination of the coefficient estimates in table 7 indicates that there
are systematic differences in the values for test and control samples. We
estimate the differences in coefficients for matched pairs of test and control
firms. The sample distributions of these differences are reported in table 7. A
Z-test, described earlier in section 3.2, is used to test the significance of the
mean coefficient differences. The test firms have a significantly higher com-
pensation-earnings elasticity than the control firms. Subsequent to the
accounting change, the test firms experience an increase in fixed compensation
and a decrease in their compensation-earnings elasticity relative to the control
firms. However, these differences are statistically insignificant. These findings
indicate that (1) firms that change to LIFO have higher compensation-earn-
ings elasticities than firms that do not change their inventory method, and (2)
the changes in model parameters are not systematically related to the account-
ing method change.
Results of tests of whether the structural change in the compensation model
offsets the effect of the accounting change on earnings-related remuneration
are reported in table 8. The cross-sectional mean (median) of the average
percent prediction error for compensation is 9.5 (10.5) percent for the test
sample. The t-statistic, that evaluates the significance of the mean, is signih-
cantly different from zero at the five percent level. The mean (median) for the
control sample is 8.3 (11) percent. The t-statistic on the mean is also signifi-
cantly different from zero. These results indicate that on average there is an
economy- or industry-related increase in compensation of eight to eleven
percent that coincides with the period following the LIFO changes, the late
1970s and early 1980s. The test firm CEOs have marginally larger average
percentage increases in salary and bonus compensation during these same
years. However, the t-statistic evaluating the difference in these means is not
significant. The results therefore support the null hypothesis that the com-
pensation committee adjusts the model parameters to offset the effect of the
accounting change.
P. Heu!v, S. Kong und K. Palepu. Accounting procedure changes and CEO compensution 21
Table 8
Summary of average percentage prediction errors for compensation in years subsequent to a
change to the LIFO inventory method.a
“Percentage prediction errors are the percentage difference between actual compensation and
compensation that would have been paid in the absence of a LIFO change and a model change
[see eq. (7)]. The results are for the cross-sectional distribution of time-series estimates of mean
percentage prediction errors for 50 test firms and 50 control firms. Each time-series contains
between six and eleven observations.
‘The test sample comprises 50 firms that changed to LIFO in the period 1970-1976. Each test
firm has a matched control firm with the same 2-digit SIC code and with no inventory or
depreciation method changes ten years before and after the year of the test firm’s method change.
Control tirms are found for 50 of 52 inventory test firms.
‘Significant at the five percent level using a two-tailed test.
IsThis is a sizeable decrease in salan, and bonus that far exceeds the effect of the inventory
change in subsequent years’ compensation. We therefore test whether the compensation commit-
tee adjusts reported earnings for the effect of the accounting change in only the year of the change
rather than in all following years. We estimate eq. (2) with the explanatory variable
ln[ REA RN,/AEA RN,] being set to zero in all years other than the year of the change to LIFO.
The coefficient is positive and significant, implying that the compensation committee does not
adjust compensation for the effect of the inventory method change even in the year of the change.
28 P. Heu!v, S. Kmg und K. Palepu, Accountingprocedure chunges and CEO compensation
Table 9
Summary statistics for compensation effects of inventory method change as a percentage of actual
compensation for the year of the inventory change and the subsequent ten years.”
Table 10
Summary of estimated coefficients for regression tests of the hypothesis that the parameters of the
compensation-earnings relation change following a change to the straight-line depreciation
method.a
ln(COMP,)=~a,D,,+P,ln(REARN,)+~2DI/M,+P,DUM,ln(REARN,)+u,.h
,=I
Test sumple’
Control sample’
“These results are for the cross-sectional distribution of time-series regressions for 37 test lirms
and 37 control firms. Each time series contains between 14 and 21 observations.
bCOMP, = salary + bonus for CEO in year 1; REARN, = reported earnings in year t: D,, = 1 if
individual i is CEO of the firm in year t and 0 otherwise; n = number of individuals who hold the
CEO position in the sample period; and DUM, = 0 prior to the year of the depreciation change
and 1 thereafter.
‘The test sample comprises 37 firms that changed to straight-line depreciation in the period
1967-1974. Each test firm has a matched control firm with the same 2-digit SIC code and with no
inventory or depreciation method changes ten years before and after the year of the firm’s method
change. Control firms are found for 37 of 38 depreciation test firms,
dThese distributional statistics are for the differences in coefficients between matched pairs of
test and control firms. The Z-statistic tests the significance of the sample mean difference for each
cofficient.
‘Significant at the five percent level using a two-tailed test.
‘Significant at the one percent level using a two-tailed test.
sSignificant at the one percent level using a one-tailed test.
JHC R
30 P. Hea@, S. Kang and K. Palepu, Accounfingprocedure chmges and CEO compensut~on
Table 11
Summary of average percentage prediction errors for compensation in years subsequent to a
change to the straight-line depreciation method.”
Difference between
Test Control test and control
sample’ sample’ samples
“Percentage prediction errors are the percentage difference between actual compensation and
compensation that would have been paid in the absence of a straight-line depreciation change and
a model change [see eq. (7)]. The results are for the cross-sectional distribution of time-series
estimates of mean percentage prediction errors for 37 test firms and 37 control firms. Each
time-series contains between six and eleven observations.
bThe test sample comprises 37 firms that changed to straight-line depreciation in the period
1967-1974. Each test firm has a matched control firm with the same 2-digit SIC code and with no
inventory or depreciation method changes ten years before and after the year of the test firm’s
method change. Control firms are found for 37 of 38 depreciation test firms.
‘Significant at the five percent level using a two-tailed test.
Table 12
Summary statistics for compensation effects of depreciation method change as a percentage of
actual compensation for the year of the depreciation change and the subsequent ten years.”
Once again, we believe that an equally plausible explanation for the above
finding is that the magnitude of the effect of the accounting change on
compensation is small relative to the effect of the time-dependent change in
compensation, preventing us from discriminating between the null and alter-
native hypotheses. Table 12 reports the distribution of the percentage effect of
the depreciation change on bonus and salary awards assuming that the
compensation committee does not adjust the earnings definition or the param-
eters of the compensation-earnings relation. The mean (median) percentage
increase in salary and bonus awards in the year of the depreciation change is
2.4 (0.8) percent. A time-series average of the compensation effect of the
change is estimated for each firm for years 0 to 10 and years 1 to 10. The
cross-sectional mean (median) of these averages is 2.7 (1.4) percent for years 0
to 10 as well as for years 1 to 10.
lhSee Ball (1985) and Demski and Sappington (1985) for a discussion of selection of accounting
procedures when managers have inside information on the costs and benefits of alternative
accounting and production/investment decisions.
“There may be several reasons for this. The potential effect of the LIFO change on compensa-
tion may be overstated if firms in our sample have bonus plans with binding upper limits on
awards in the year of the inventory method change. As Biddle and Lindahl (1982) and Ricks
(1982) document, firms change to LIFO in years of large earnings increases. If there is an upper
limit on bonus awards, it is likely to be binding in these years under both FIFO and LIFO. Healy
(1985) finds that 46 percent of Fortune 250 industrial firms with usable bonus plans have an upper
limit on the pool of funds available for bonus awards. Also, managers own stock in the companies
that employ them. Their portfolio wealth will therefore i&crease following a change in LIFO if
equity prices reflect the accompanying tax savings. Finally, executives’ human capital is likely to
increase following a change to LIFO since the decision benefits shareholders.
34 P. He+, S. Kung und K. Palepy Accountingprocedure changes and CEO compensation
References
Abdel-khalik, A.R., 1985, The effect of LIFO-switching and firm ownership on executives’ pay,
Journal of Accounting Research 23, 421-447.
Ball, Ray, 1985, Accounting, auditing and the nature of the firm, Unpublished paper (Australian
Graduate School of Management, and University of Chicago, Chicago, IL).
Biddle, G.C. and F.W. Lindahl, 1982, Stock price reactions to LIFO adoptions: The association
between excess returns and LIFO tax savings, Journal of Accounting Research 20, 551-588.
Boyes, W. and D. Schlagenhauf, 1979, Managerial incentives and the specification of functional
forms, Southern Economics Journal 45, 1225-1232.
Christie, Andrew, 1986, Aggregation of test statistics: An evaluation of the consistency of evidence
on contracting and size hypotheses, Unpublished paper (University of Southern California, Los
Angeles, CA).
Demski, Joel S. and David E. Sappington, 1985, Accounting procedures in a principal-agent
model, Unpublished paper (Stanford University, Stanford, CA, and University of Penn-
sylvania, Philadelphia, PA).
Fox, Harland, 1980, Top executive compensation (The Conference Board, New York).
Hagerman, R.L. and M.E. Zmijewski, 1979, Some economic determinants of accounting policy
choice, Journal of Accounting and Economics 1, 141-162.
Healy, P.M., 1985, The effect of bonus schemes on accounting decisions, Journal of Accounting
and Economics 7, 85-107.
Holthausen, R.W., 1981, Evidence on the effect of bond covenants and management compensa-
tion contracts on the choice of accounting techniques: The case of the depreciation switch-back,
Journal of Accounting and Economics 3, 73-109.
Holthausen, R.W. and R.W. Leftwich, 1983, The economic consequences of accounting choice:
Implications of costly contracting and monitoring, Journal of Accounting and Economics 5,
77-117.
Kaplan, R. and R. Roll, 1972, Investor evaluation of accounting information: Some empirical
evidence, Journal of Business 45. 225-257.
Larcker, David F., 1983, The association between performance plan adoption and corporate
capital investment, Journal of Accounting and Economics 5, 3-30.
Murphy, K., 1985, Corporate performance and managerial remuneration: An empirical analysis,
Journal of Accounting and Economics 7, 11-42.
Ricks, W., 1982, The market’s response to the 1974 LIFO adoptions, Journal of Accounting
Research 20, 367-387.
Siegel, Sidney, 1956, Nonparametric statistics for the behavioral sciences (McGraw-Hill, New
York).
Smith, Clifford and Ross Watts, 1982, Incentive and tax effects of executive compensation plans,
Australian Journal of Management 7, 139-157.
Sunder, Shyam, 1973, Relationship between accounting changes and stock prices: Problems of
measurement and some empirical evidence, in: Empirical research in accounting: Selected
studies 1973, Supplement to Vol. 11 of Journal of Accounting Research, l-45.
Sunder, Shyam, 1975, Stock price and risk related to accounting changes in inventory valuation,
Accounting Review 50, 305-315.
Thiel, H., 1971, Principles of econometrics (Wiley, New York).
Watts, Ross and Jerold Zimmerman, 1986, Positive accounting theory (Prentice Hall, Englewood
Cliffs, NJ).